Sam Altman on GPT-6: Why AI Needs Memory

Sam Altman, CEO of OpenAI, recently spoke about the future of artificial intelligence and suggested that the GPT-6 model will arrive sooner than expected. In an interview in San Francisco, he emphasized that GPT-6 will not merely answer questions but will adapt to users. This means it will remember their preferences, daily routines, and personal quirks, enabling more natural interactions.

Altman stated clearly: “People want memory. People want product features that require us to be able to understand them.” According to him, memory is the key to making ChatGPT a truly personal tool. OpenAI is working with psychologists to monitor how people feel when using such AI over the long term and to measure their well-being over time. This data is not yet public, but Altman suggested that it could be released.

Problems with GPT-5 and Lessons for the Next Version

The launch of GPT-5 was not ideal. Users on social media complained that it was colder, less connected, and less useful than the previous version. Altman acknowledged this and said: “I like the new one much better.” OpenAI quietly updated GPT-5’s tone to make it “much warmer.” Altman emphasized that GPT-6 will arrive sooner than the interval between GPT-4 and GPT-5 and will differ by focusing on adapting to the user.

Altman also mentioned that future versions of ChatGPT will allow users to define its tone and personality. The model is intended to be ideologically neutral, in line with a recent order from the Donald Trump administration requiring AI systems used by the federal government to be neutral and customizable. Altman said: “I think our product should have a fairly centrist, balanced position, and then you should be able to move it quite far.” If a user wants the model to be very progressive, it should allow that, just as it should allow a conservative setting.

Privacy Concerns

Altman called improved memory his favorite feature of the year but warned about privacy issues. Temporary memory in current models is not encrypted, meaning sensitive information could be at risk. Altman confirmed that encryption “very well could be” added, though he gave no specific timeline.

Stronger protection is especially necessary for queries involving legal or health information. Altman said: “It is in society’s interest for people to receive good health advice... good legal advice. And if you can get better versions of that from AI, you should have the same protection as when you get it from a doctor or lawyer.”

Additional information from recent reports confirms that GPT-6 is expected to introduce memory that will retain users’ preferences, conversation history, tone, and even emotional context. This should help it move from a “smart chatbot” to something more like a companion that builds trust and continuity. OpenAI is studying the psychological effects, including the risks of emotional dependence and “artificial intimacy.” Altman emphasized that development is being guided by feedback from users who do not want to start every conversation from scratch.